A methodological framework for assessing and reducing temporal uncertainty in paleovegetation mapping from late-Quaternary pollen records

Mapping past vegetation dynamics from heterogeneous databases of fossil-pollen records must face the challenge of temporal uncertainty. The growing collection of densely sampled fossil-pollen records with accurate and precise chronologies allows us to develop new methods to assess and reduce this uncertainty. Here, we test our methods in the context of vegetation changes in eastern North America during the abrupt climate changes of the last deglaciation. We use the network of fossil-pollen records in the Neotoma Paleoecology Database (www.neotomadb.org) and data contributed by individual investigators. Because many of these records were collected decades before the current generation of ¹⁴C and age-model technologies, we first developed a framework to assess the overall reliability of ¹⁴C chronologies by systematically evaluating individual ¹⁴C ages and associated chronologies. We developed a qualitative ranking scheme for individual ¹⁴C ages that combines information about their accuracy and precision. 'Benchmark' pollen records were defined to have at least one ¹⁴C age with an accuracy within 250 years and a precision less than 500 years that is within 1000 years of the time interval of interest, and at least five pollen samples per 1000 years across this time period. Only 22 of >350 late-Pleistocene pollen cores in eastern North America met the benchmark criteria. We then used Bayesian change-point analysis to identify widespread ecological events (Picea decline, Quercus rise, and Alnus decline), and interpolated the ages of these events from the benchmark sites to non-benchmark sites. Leave-one-out cross-validation analyses with the benchmark sites indicated that the spatial error associated with interpolation was less for inverse distance-weighting (IDW) than thin-plate splines (TPS) and was about 500 years for the three biotic events. By comparison, the difference between the original ages of events at poorly constrained sites and the biostratigraphic ages interpolated from the benchmark sites was close to 1000 years, suggesting that the use of biostratigraphic ages can significantly improve the age models for poorly constrained sites. Overall, these analyses suggest that the temporal resolution of multi-site syntheses of late-Pleistocene fossil-pollen data in eastern North America is about 500 years, a resolution that allows analysis of ecological responses to millennial-scale climate change during the last deglaciation.